Turn on ideal line (top row number 4) to see where the AI are trying to go
This isn't an open config, it's Fern Bay Green with AutoX objects placed. You are right in that AI cannot drive open configs, but this is legitimate as far as the race start logic is concerned.
I think this is a bug. The race should not be able to be started with AI, and when any AutoX objects are placed, not just a start position.
Thanks, I updated my post just as you wrote that by the looks of it. A schoolboy error methinks.
Very interesting. I was reading through that thread, and somehow missed EQ Worry's "useful post"
My experiment has finished, and the ideal line seems to be set in stone (read pth) I cannot detect any differences either between the visible ideal lines for AI with different setups, nor different cars. The only file updated is the .trs in the knw folder.
I surmise that the ideal line is probably a best fit curve for the shortest path between intersections of path nodes and road limits, plus some other factor that accounts for velocity perhaps, or even between the node centre and road limit on the shortest path.
How did you generate the ideal line in your images PMD9409? Are they a merge of the originals? It looks like they get a bit wobbly where the different configurations join.
I've searched for possible merged pth files and found non. Did Eq Worry send them to you off forum? Would you care to share?
I've been looking at the .pth files and I'm having trouble already. Firstly there only appear to be 11 in the pth folder. AS1-7, KY1-3, and WE1.The pth.txt published for patch Y doesn't seem to describe what I'm seeing. Anyone know if its valid for the current format? According to the header, offsets 6 and 7 contain the version and revision respectively, but I see 00 and fa. That's ok, but pth.txt says "0 - do not read file if > 0" for both entries. Also the node count at offset 8 is zero. I'm thoroughly confused by it now. Any tips would be welcomed.
Dygear, I think this might be what Cargame refers to.
[Edit]
Doh!
I'm looking at the pth folder, and I guess they are compiled. Offset 0 should read "LFSPTH" not "SRPATH" Look at the .pth files in SMX_PTH_S2Y.zip instead. In other words ignore my previous comments.
A sample merged .pth file would be nice to work with though.
[/edit]
You have the wrong end of the stick. BL rally sections are still there for open configs. The only differences are that barriers along the sides of those sections are back. This means you cannot drive into the lake for example.
There is no obligation to download any of these test patches. They are exactly that - for testing and bug reports.
And the full quote giving an explanation just for reference
I'm guessing avetere generated his svg using that data, and using interpolation a new path could be written out. Another thing that strikes me, is will an open config layout even attempt to read a path file? The existing ones are all named per existing track configuration, and we only have the [x,y] suffix which does not indicate a particular route. It would also need to depend on the layout file for which there has never been a path association.
Very good idea. I was just thinking about layouts. There would need to be a valid/verified pth for each and every layout wouldn't there? I'm guessing that Scawen has never released, or intended to release a path editor because the paths for the official configurations were all that were needed. Any autocross layout - as I imagine was originally intended for only autox - would have been beyond the original scope. That has all changed now if open configs are here to stay.
Yes the existing node data should still be relevant, it should only differ in its direction and where they fall in between existing nodes, they could be interpolated. As far as track width is concerned, can't this be extracted from the smx files?
[Edit]
I just had a horrible thought.
Cruise servers would like to be able to drive both ways on a section of track. If the existing nodes widths (not inter node distance) are used, there would be a conflict. This is also true for rally style layouts that may use the same bit of track, so some method of defining the width would be needed too. Not something that is too pressing right now, as the paths - any paths - need to be created first, but worth bearing in mind. [/edit]
Last edited by Squelch, .
Reason : Added more conjecture
I believe you are correct. I thought about it after I read back what I said before.
Good find! That sort of backs up the theory that a car was used to set the initial paths, but as you say, which one? The spreading out of the nodes in avetere's svg also suggests that they were placed/recorded at different speeds too, so the reasonable speed may have just been different for each configuration.
That quote just about confirms it.
So, now to find the secret sauce that allows recording/updating paths. I'm still running my empirical test just to see if something crops up and some correlation can be made.
That is a good question. If for example the path is recorded using a UF1, then because of the lower speeds, the resolution would be higher. This by implication would mean all faster cars would be able to use this path.
I'm testing that theory right now. There is an entry under options/misc - Update path Off/User Car/All Cars that I don't quite understand. Does this mean the path is dynamic? Anyway I have an AI with a bad setup driving Aston right now. Assuming the AI tries to drive the ideal line whenever possible, I'm letting it do its thing, and do not see any obvious change in ideal line - yet. The bad setup means it cannot keep a good line, and my hope is once done, and a restart performed, There will be a different ideal line, and by implication, an updated path. I'll compare screen shots of a few corners at different points and between the sessions, as well as run a diff on the files to look for changes.
I see what you mean now. However, I have this feeling, and from various comments Scawen has made over the years about his use of AI drivers, that the paths could generated dynamically from a reference AI car. If the track was driven once, or a standard set of path nodes generated automatically, then letting the AI learn and record a new path seems plausible don't you think?
I was wondering that, and perhaps my experiment will show something up along these lines. It may well be that the ideal line is a shortest path bounded by the track limits and a best fit between along the path nodes.
I watched this the other day and thoroughly enjoyed it. Bad Boy Bubby
It is typical Australian cinema. Disturbing, moving, funny, and totally captivating. One nice quirk is that virtually all of the sound was recorded from the main actors (Nicholas Hope) perspective using two microphones hidden behind his ears.
Just to float some ideas; The paths that avetere shows in his SVG image seem to differ in position only, and may be from the relative velocities of the vehicle that possibly recorded them. Would a matching algorithm work for finding the nearest node in the direction of travel work? Dygear's 2nd suggestion could be combined with this to reuse the most appropriate path for any given configuration.
I'm not sure how the ideal line is generated, but it isn't just a spline fit. I'm guessing the velocity/mass of the vehicle is also fitted to this curve.
[Edit]
Whiskey makes a point that backs up my observation.
The difference in positions - take the northernmost bend for example - seem to indicate they were "sampled" at regular intervals. At the apex the speed would be about the same for all 3 regular configurations that use that bend, and they all pretty much overlap. Where they stretch out may be down to the setup of the recording vehicle (different gear ratios?) if that is indeed how they were generated.
Like Whiskey, my observations need a disclaimer too.
[/edit]
Last edited by Squelch, .
Reason : Included Whiskey's reply
Quite some time back this was talked about (LFS spectate?). My suggestion back then was to encapsulate a replay file as a broadcast rtp stream.
I've experimented with VideoLan that can render directx to a transport stream from a client connected as spectator. This only takes one slot on the server, and can be done on another remote server to feed the live stream. I had mixed results however, and perhaps someone with better kit than me might have more luck.
Using a stream there would be no LFS connection other than the official spectator.
Same as above.
Can be provided for via IM server.
Not needed if using the official one spectator slot.
Server admins can decide not to allow the streamed spectator.
I totally agree with this last point. I have long wished for a method of watching the big races without having to take up a slot, or having to wait for edited highlights.
The streamed method would work on services like Ustream, which also has a chat client attached. Another benefit of this method would be that the spectating streaming server could also be human controlled.
Anyone fancy trying thier hand at race coverage directing?
[Edit]
The old thread is LFS TV and I seem to recall an even older one that was on RSC "Live to Spectate"
I must have driven that corner a hundred times now to try and replicate the issue. Every time I'm on the throttle later than Spiderbait, It's just a difference in driving style I suppose.
Can anyone else replicate this?
Last edited by Squelch, .
Reason : name correction
I've just bee analysing your SPR and there seems to be a texture join at the point the car gets launched. The previous reports also seem to happen around similar points, so my earlier theory might still hold water.
Image sequence
The front wheel force shows the start of a bump in road surface. Texture join highlighted.
Rear wheel forces for same bump (closer). Join just above highlight.
Rear wheel just after the join (highlighted) and at point where the car is "launched"
[edit]I just checked my replay, and I don't accelerate so soon which might explain why I can't replicate. In image 2 there are forward acceleration forces just before the "bump" which immediately disappear at the join. Later frames show a negative force on the rear wheels as the car is launched [/edit]
I think you are referring to the phantom collisions that have been reported during multiplayer sessions using the latest patch sets.
Scawen has said it concerns him, and I believe there are too many occurrences of this to ignore now. What you propose is an excellent idea, as all data on the local machine would be recorded, and not just the car position that simply shows the outcome in an MPR.
Looking at the MPR's for those reports does not reveal any significant forces acting on the car at the time of the incident. Perhaps a local SPR would show more of what is happening.
There could be another important benefit other than this bug hunt. That is a player could view live how a car behaves while slip streaming which is difficult to interpret from a RAF. This could greatly aid set up tweaks.
A simple test would be to see if the delta is more than 5 seconds below the average time set. There might be a legitimate hotlap set, but at least a potentially bogus one would be flagged and investigated.
BTW, I don't see that hotlap on LFSW, so I guess it's been removed.
[Edit] message to self. refresh page before reply - thanks Scawen/Vic [/Edit]
Thank you for that one in particular as well as the rest of the fixes.
Hehe, what a dilemma. It seems Scawen is damned if if he does, and damned if he doesn't. I like the fact we could use some off road sections between the barriers in exactly this manner. Some of the reports of missing barriers however, have been just for the sake of it I feel (lake?), but in many cases they have been legitimate places where a mistake could lead to passing through scenery into the netherworld.
I still like my tape idea for those areas that need marking off, but don't necessarily need a physical barrier.
I use this tool for checking, but there are others out there.
A quick request: Can checksums be published please? I still don't know how or why I ended up with a broken installation, but being able to check against the checksums would have helped.
It appears it was a broken update. I updated z31 to z32, and although it looked correct, I ran into problems while bug hunting. A fresh install of z28 and updating to z32 fixed the problem.
I'm posting this feedback so that it might be considered should any one else run into similar problems in future.
The replay is paused and I enter Shift+U and set follow.
Stepping forwards is no problem, but stepping back [index 0:27] the view is reset.
I return to the car, and follow it again but this time locking the view [index 0:44].
Stepping forwards is again without problems, and stepping backwards now stays with the car [index 0:50]
I then turn off the lock and the camera is once again reset when backstepping [index 1.04]
The last part of the video shows what happens when I do not return to the car, but set follow mode from the reset camera position. Stepping forward works as expected, but when I backstep, the view is not only reset, but increases in distance from the car. Repeating this (using reset camera position and frame stepping) the camera gets further and further from the car. [indices 1:17 and 1:24] This later part is new to me (I didn't keep Z30/1 so cannot check back).
Is this the desired behaviour?
[Additional info]
The replay is a saved SPR made with Z32.
Method of updating was to copy and overwrite contents of Z31 from zip. Paths were maintained.
This last point is the only thing I can think of that might be different. Should I have updated from Z28?
[/Additional]
Last edited by Squelch, .
Reason : Additional info
Basically it this happens when the replay is in pause mode. My recipe should have read
6.5 Follow car again
The follow cam works while stepping forward, but the moment you step backwards the view is reset to a static view at the point the original follow car command was invoked. Locking toggle the view seems to "set" the point at which the view is reset to (and this includes a non static follow cam) I can confirm that the car is followed after a replay restarts (which it didn't do before) it just seems paused rewind is the problem now.
The only place I didn't drive.
I think the bridge is intersecting the fence at that point. The fence is solid all the way before and after that point up until the stands where you fall through the grass.
I see the same. It is only at a long distance that the z-buffering fight seems to happen. This is also seen in other game titles btw.
Scawen, the follow car is still exhibiting lost focus on rewind. Locking the camera seems to have the same workaround.
Steps for replication.
Start replay
Enter Shift+U
Follow car
Exit Shift+U
Pause replay
Enter Shift+U
Step backwards (< key)
You find that the view is reset to the original follow car position. Locking the view still has the effect of setting the camera, but subsequent changes to view are lost without toggling the lock.
PS. We need another name for Shift+U - free camera perhaps?
Hmm, that seems odd. The fences are solid in the position you say. The only problem I had was falling through the grass in front of the stands while driving along the fence, but that hasn't been addressed in this patch, and I consider this area to be no drive anyhow.